Deprecate `upcast_vae` in SDXL based pipelines (#12619)

* update

* update

* Revert "update"

This reverts commit 73906381ab76da96eb8f9b841177cd4f49861eb1.

* Revert "update"

This reverts commit 21a03f93ef0fbfa5f7a7d97708f75149b1d1b3b0.

* update

* update

* update

* update

* update

src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_instruct_pix2pix.py

@@ -22,11 +22,6 @@ from transformers import CLIPTextModel, CLIPTextModelWithProjection, CLIPTokeniz
 from ...image_processor import PipelineImageInput, VaeImageProcessor
 from ...loaders import FromSingleFileMixin, StableDiffusionXLLoraLoaderMixin, TextualInversionLoaderMixin
 from ...models import AutoencoderKL, UNet2DConditionModel
-from ...models.attention_processor import (
-    AttnProcessor2_0,
-    FusedAttnProcessor2_0,
-    XFormersAttnProcessor,
-)
 from ...models.lora import adjust_lora_scale_text_encoder
 from ...schedulers import KarrasDiffusionSchedulers
 from ...utils import (
@@ -590,22 +585,12 @@ class StableDiffusionXLInstructPix2PixPipeline(
 
     # Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl.StableDiffusionXLPipeline.upcast_vae
     def upcast_vae(self):
-        dtype = self.vae.dtype
-        self.vae.to(dtype=torch.float32)
-        use_torch_2_0_or_xformers = isinstance(
-            self.vae.decoder.mid_block.attentions[0].processor,
-            (
-                AttnProcessor2_0,
-                XFormersAttnProcessor,
-                FusedAttnProcessor2_0,
-            ),
+        deprecate(
+            "upcast_vae",
+            "1.0.0",
+            "`upcast_vae` is deprecated. Please use `pipe.vae.to(torch.float32)`. For more details, please refer to: https://github.com/huggingface/diffusers/pull/12619#issue-3606633695.",
         )
-        # if xformers or torch_2_0 is used attention block does not need
-        # to be in float32 which can save lots of memory
-        if use_torch_2_0_or_xformers:
-            self.vae.post_quant_conv.to(dtype)
-            self.vae.decoder.conv_in.to(dtype)
-            self.vae.decoder.mid_block.to(dtype)
+        self.vae.to(dtype=torch.float32)
 
     @torch.no_grad()
     @replace_example_docstring(EXAMPLE_DOC_STRING)

src/diffusers/pipelines/t2i_adapter/pipeline_stable_diffusion_xl_adapter.py

@@ -34,15 +34,12 @@ from ...loaders import (
     TextualInversionLoaderMixin,
 )
 from ...models import AutoencoderKL, ImageProjection, MultiAdapter, T2IAdapter, UNet2DConditionModel
-from ...models.attention_processor import (
-    AttnProcessor2_0,
-    XFormersAttnProcessor,
-)
 from ...models.lora import adjust_lora_scale_text_encoder
 from ...schedulers import KarrasDiffusionSchedulers
 from ...utils import (
     PIL_INTERPOLATION,
     USE_PEFT_BACKEND,
+    deprecate,
     is_torch_xla_available,
     logging,
     replace_example_docstring,
@@ -779,21 +776,12 @@ class StableDiffusionXLAdapterPipeline(
 
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_upscale.StableDiffusionUpscalePipeline.upcast_vae
     def upcast_vae(self):
-        dtype = self.vae.dtype
-        self.vae.to(dtype=torch.float32)
-        use_torch_2_0_or_xformers = isinstance(
-            self.vae.decoder.mid_block.attentions[0].processor,
-            (
-                AttnProcessor2_0,
-                XFormersAttnProcessor,
-            ),
+        deprecate(
+            "upcast_vae",
+            "1.0.0",
+            "`upcast_vae` is deprecated. Please use `pipe.vae.to(torch.float32)`. For more details, please refer to: https://github.com/huggingface/diffusers/pull/12619#issue-3606633695.",
         )
-        # if xformers or torch_2_0 is used attention block does not need
-        # to be in float32 which can save lots of memory
-        if use_torch_2_0_or_xformers:
-            self.vae.post_quant_conv.to(dtype)
-            self.vae.decoder.conv_in.to(dtype)
-            self.vae.decoder.mid_block.to(dtype)
+        self.vae.to(dtype=torch.float32)
 
     # Copied from diffusers.pipelines.t2i_adapter.pipeline_stable_diffusion_adapter.StableDiffusionAdapterPipeline._default_height_width
     def _default_height_width(self, height, width, image):

src/diffusers/pipelines/text_to_video_synthesis/pipeline_text_to_video_zero_sdxl.py

@@ -19,16 +19,12 @@ from transformers import (
 from ...image_processor import VaeImageProcessor
 from ...loaders import StableDiffusionXLLoraLoaderMixin, TextualInversionLoaderMixin
 from ...models import AutoencoderKL, UNet2DConditionModel
-from ...models.attention_processor import (
-    AttnProcessor2_0,
-    FusedAttnProcessor2_0,
-    XFormersAttnProcessor,
-)
 from ...models.lora import adjust_lora_scale_text_encoder
 from ...schedulers import KarrasDiffusionSchedulers
 from ...utils import (
     USE_PEFT_BACKEND,
     BaseOutput,
+    deprecate,
     is_invisible_watermark_available,
     logging,
     scale_lora_layers,
@@ -457,22 +453,12 @@ class TextToVideoZeroSDXLPipeline(
 
     # Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl.StableDiffusionXLPipeline.upcast_vae
     def upcast_vae(self):
-        dtype = self.vae.dtype
-        self.vae.to(dtype=torch.float32)
-        use_torch_2_0_or_xformers = isinstance(
-            self.vae.decoder.mid_block.attentions[0].processor,
-            (
-                AttnProcessor2_0,
-                XFormersAttnProcessor,
-                FusedAttnProcessor2_0,
-            ),
+        deprecate(
+            "upcast_vae",
+            "1.0.0",
+            "`upcast_vae` is deprecated. Please use `pipe.vae.to(torch.float32)`. For more details, please refer to: https://github.com/huggingface/diffusers/pull/12619#issue-3606633695.",
         )
-        # if xformers or torch_2_0 is used attention block does not need
-        # to be in float32 which can save lots of memory
-        if use_torch_2_0_or_xformers:
-            self.vae.post_quant_conv.to(dtype)
-            self.vae.decoder.conv_in.to(dtype)
-            self.vae.decoder.mid_block.to(dtype)
+        self.vae.to(dtype=torch.float32)
 
     # Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl.StableDiffusionXLPipeline._get_add_time_ids
     def _get_add_time_ids(